A principal objective of the welfare reform act of 1996 (PRWORA) was to encourage welfare recipients to obtain jobs rapidly, a strategy termed "Work First." Much analysis shows that Work First raises the incidence of direct-hire and—in a sizable minority of cases—temporary-help agency jobs among welfare clients. But the effect of these jobs on longer term labor market outcomes, such as labor force participation, earnings, and welfare recidivism, is unknown. Because welfare recipients who obtain jobs rapidly are positively selected from the pool of all Work First participants, a simple comparison of long-term outcomes among job takers and non-takers is potentially misleading. We evaluate the effects of Work First job placements on program recidivism and earnings of welfare recipients over a two-year period following Work First program assignment. We draw upon administrative data from an unusual policy experiment in the state of Michigan. Welfare recipients in a major metropolitan were randomly assigned to a large number of Work First contractors that had widely varying placement rates in direct-hire and temporary-help jobs but provided otherwise similar services. These assignments significantly impacted job-finding rates among ex ante comparable clients living in the same neighborhoods. We find that Work First placements in direct-hire jobs substantially raised earnings over the two years following program assignment, but that temporary help agency placements yielded no lasting gains in welfare recipients' employment and earnings.